OCI AI Foundations study plan

Define supervised, unsupervised, and reinforcement learning, then classify 10 given real-world scenarios into the correct paradigm with a written justification for each.

Distinguish classification from regression and select the appropriate evaluation metric (accuracy, precision, recall, F1, ROC-AUC, RMSE, MAE) for 8 sample problem statements.

Explain the bias-variance tradeoff and diagnose overfitting versus underfitting from 5 written descriptions of training and validation curves.

Compute precision, recall, specificity, and F1 by hand from a supplied 2x2 confusion matrix, then interpret which metric matters most for an imbalanced fraud-style example.

Sequence the ML project lifecycle stages (problem framing, data collection, feature engineering, training, evaluation, deployment, monitoring) and map each stage to the risks that arise there.

Describe data preparation techniques including normalization, handling missing values, class imbalance strategies, and train/validation/test splitting, and apply the correct technique to 6 short scenarios.

Differentiate batch versus real-time inference and choose the right pattern for 5 stated workloads.

Contrast OCI AI Services, OCI Data Science, and OCI AI Infrastructure on abstraction level, target user, and customization effort, then place 12 example use cases on the correct tier.

List the core capabilities of OCI Language (sentiment analysis, named entity recognition, key phrase extraction, text classification, language detection, translation) and select the correct capability for 8 given text-processing requirements.

List the core capabilities of OCI Vision (image classification, object detection, OCR) and differentiate custom-model training from pretrained-feature use in 5 scenarios.

Compare OCI Document Understanding against raw Vision OCR by identifying which handles key-value extraction, table extraction, and document classification.

Write a written decision procedure that takes a plain-English requirement and outputs the correct OCI AI service plus the input/output data type.

Distinguish OCI Speech transcription from speech synthesis and identify supported input and output formats for each.

Explain OCI Anomaly Detection training modes and data requirements, then match 6 monitoring scenarios to the correct configuration.

Explain OCI Forecasting inputs, forecast horizon, and quantile output, and choose it over regression for 4 time-series scenarios.

Diagnose 10 mixed service-selection scenarios spanning Language, Vision, Speech, Document Understanding, Anomaly Detection, and Forecasting, and record the reason for each miss.

Explain tokenization, embeddings, attention, and the encoder-decoder structure at a conceptual level, and identify the purpose of each in a transformer-based LLM.

Differentiate pretraining, fine-tuning, and retrieval-augmented generation, and choose the correct approach for 8 stated business constraints (cost, data freshness, proprietary knowledge, latency).

Author 6 prompts that each apply at least one named technique (role prompting, few-shot examples, chain-of-thought, output-format specification, guardrails) and iterate each one at least once.

Identify sources of hallucination and describe at least 4 mitigation techniques including grounding, citations, temperature control, and output validation.

Label the components of a feed-forward neural network (input layer, hidden layers, weights, biases, activation, loss, optimizer) and describe forward propagation and backpropagation in sequence without notes.

Compare CNN, RNN/LSTM, and transformer architectures and assign the better architecture to 6 stated task types (image classification, sequence forecasting, translation, tabular prediction).

Explain the effect of activation function choice, dropout, learning rate, batch size, and epoch count on convergence, and predict the symptom of 5 misconfigurations.

Explain why GPUs accelerate deep learning and identify OCI GPU shapes and cluster networking options at a conceptual level.

Describe OCI Data Science components (notebook sessions, jobs, conda environments, model catalog, model deployment, pipelines) and map each to the lifecycle stage it supports.

Describe OCI AI Infrastructure building blocks including GPU bare metal shapes, cluster networking, and high-performance storage, and match 5 workloads to an appropriate infrastructure pattern.

Identify responsible AI risks (bias, fairness, transparency, explainability, privacy, human oversight) and propose one mitigation for each in a 6-item checklist.

Explain how OCI security controls (IAM policies, compartments, Vault, encryption at rest and in transit) apply to AI workloads and data.

Re-derive from memory the one-page summaries for all six domains and flag every gap found.

Complete two full-length timed practice exams under exam conditions and log every incorrect answer by domain.

Re-study only the three weakest subtopics identified from the exam log and rewrite each as a corrected rule.

Recite the OCI AI service-selection procedure and the RAG-versus-fine-tuning decision criteria without reference material.